Transcript Ch2 - University of Central Oklahoma

Signals and Systems
Chapter 2
Biomedical Engineering
Dr. Mohamed Bingabr
University of Central Oklahoma
Outline
• Signals
• Systems
• The Fourier Transform
• Properties of the Fourier Transform
• Transfer Function
• Circular Symmetry and the Hankel Transform
Introduction
Signal Type
- Continuous Signal: x-ray attenuation
- Discrete Signal: times of arrival of photons in a
radioactive decay process in PET
- Mixed signal: CT scan signal g(l,θk)
System Type
- Continuous-continuous system
- Continuous-discrete system
Signals
2-D continuous signal is defined as f(x,y)
(x,y) : is a pixel location
f : is pixel intensity
function
image
Point Impulse
1-D point impulse (delta, Dirac, impulse function)
𝛿 𝑥 = 0,
𝑥 ≠ 0,
𝛿 𝑥
∞
𝑓(𝑥) 𝛿 𝑥 𝑑𝑥 = 𝑓 0 .
−∞
2-D point impulse
𝛿 𝑥, 𝑦 = 0,
∞
(𝑥, 𝑦) ≠ (0, 0)
∞
𝑓(𝑥, 𝑦) 𝛿 𝑥, 𝑦 𝑑𝑥𝑑𝑦 = 𝑓 0, 0 .
−∞ −∞
Point impulse is used in the characterization of image
resolution and sampling
𝑥
Point Impulse Properties
1- Sifting property
∞
∞
𝑓(𝑥, 𝑦) 𝛿 𝑥 − 𝜉, 𝑦 − 𝜂 𝑑𝑥𝑑𝑦 = 𝑓 𝜉, 𝜂 .
−∞ −∞
We can interpret the product of a function with a point impulse
as another point impulse whose volume is equal to the value of
the function at the location of the point impulse.
2- Scaling property
1
𝛿 𝑎𝑥, 𝑏𝑦 =
𝛿(𝑥, 𝑦)
𝑎𝑏
2- Even function
𝛿 −𝑥, −𝑦 = 𝛿(𝑥, 𝑦)
Line Impulse
Line also used to assist image resolution
𝐿 𝑙, 𝜃 =
𝑥, 𝑦 |𝑥𝑐𝑜𝑠𝜃 + 𝑦𝑠𝑖𝑛𝜃 = 𝑙
This is a line whose unite normal is oriented at an
angle θ relative to the x-axis and is at distance l from
the origin in the direction of the unit normal.
The line impulse 𝛿𝑙 𝑥, 𝑦 associated with line 𝐿 𝑙, 𝜃
𝛿𝑙 𝑥, 𝑦 = 𝛿𝑙 𝑥𝑐𝑜𝑠𝜃, +𝑦𝑠𝑖𝑛𝜃 − 𝑙
Comb and Sampling Functions
Used in medical imaging production (sampling CT
image 1024 x 1024), manipulation, and storage.
∞
𝑐𝑜𝑚𝑏(𝑥) =
𝛿(𝑥 − 𝑛)
−∞
2-D comb function
∞
∞
𝑐𝑜𝑚𝑏(𝑥, 𝑦) =
𝛿(𝑥 − 𝑚, 𝑦 − 𝑛)
𝑚=−∞ 𝑛=−∞
Sampling function
∞
∞
𝛿𝑠 𝑥, 𝑦; ∆𝑥, ∆𝑦 =
𝛿(𝑥 − 𝑚∆𝑥, 𝑦 − 𝑛∆𝑦)
𝑚=−∞ 𝑛=−∞
1-D Rect and Sinc Functions
Rect function is used in medical imaging for sectioning.
𝑟𝑒𝑐𝑡(𝑥) =
1,
0,
1
for |𝑥| <
2
1
for |𝑥| >
2
Sinc function is used in medical imaging for
reconstruction.
𝑠𝑖𝑛𝜋𝑥
𝑠𝑖𝑛𝑐 𝑥 =
𝜋𝑥
2-D Rect and Sinc Functions
𝑟𝑒𝑐𝑡(𝑥, 𝑦) = 𝑟𝑒𝑐𝑡(𝑥)𝑟𝑒𝑐𝑡(𝑦)
𝑟𝑒𝑐𝑡(𝑥, 𝑦) =
1,
0,
1
1
for |𝑥| <
and |𝑦| <
2
2
1
1
for |𝑥| > and 𝑦 >
2
2
𝑠𝑖𝑛𝑐 𝑥, 𝑦 = 𝑠𝑖𝑛𝑐 𝑥 𝑠𝑖𝑛𝑐(𝑦)
1,
𝑠𝑖𝑛𝑐(𝑥, 𝑦) = sin 𝜋𝑥 sin(𝜋𝑦)
,
2
𝜋 𝑥𝑦
for 𝑥 = 𝑦 = 0
otherwise.
Exponential and Sinusoidal Signals
𝑒(𝑥, 𝑦) = 𝑒 𝑗2𝜋
𝑢0 𝑥+𝑣0 𝑦
𝑒 𝑥, 𝑦 = 𝑐𝑜𝑠 2𝜋 𝑢0 𝑥 + 𝑣0 𝑦
𝑐𝑜𝑠 2𝜋 𝑢0 𝑥 + 𝑣0 𝑦
= 0.5𝑒 𝑗2𝜋
x and y have distance
units.
u0 and v0 are the
fundamental
frequencies and their
units are the inverse
of the units of x and y.
+ 𝑗𝑠𝑖𝑛 2𝜋 𝑢0 𝑥 + 𝑣0 𝑦
𝑢0 𝑥+𝑣0 𝑦
+ 0.5𝑒 −𝑗2𝜋
𝑢0 𝑥+𝑣0 𝑦
Separable and Periodic Signals
•
A signal f(x, y) is separable if f(x, y)= f1(x) f2(y)
•
Separable signal model signal variations
independently in the x and y direction.
•
Decomposing a signal to its components f1(x) and
f2(y) might simplify signal processing.
Periodicity
A signal f(x, y) is periodic if f(x, y)= f(x+X, y) = f(x, y+Y)
X and Y are the signal periods in the x and y direction,
respectively.
Systems
A continuous system is defined as a transformer Ϩ of an
input continuous signal f(x,y) to an output continuous
signal g(x,y).
g(x, y)= Ϩ [f(x, y)]
Linear Systems
𝐾
𝐾
𝑤𝑘 𝑓𝑘 (𝑥, 𝑦) =
Ϩ
𝑘=1
𝑤𝑘 Ϩ 𝑓𝑘 (𝑥, 𝑦)
𝑘=1
Impulse Response
If we know the system response to an impulse
𝛿𝜉𝜂 𝑥, 𝑦 = 𝛿 𝑥 − 𝜉, 𝑦 − 𝜂
then with linearity we can know the system response to
any input.
ℎ 𝑥, 𝑦; 𝜉, 𝜂 = Ϩ 𝛿𝜉𝜂 𝑥, 𝑦
ℎ 𝑥, 𝑦; 𝜉, 𝜂 is the system impulse response function
or known as point spread function (PSF).
Impulse Response
System output g() for any input f().
∞
𝑔 𝑥, 𝑦 =
∞
𝑓 𝜉, 𝜂 ℎ 𝑥, 𝑦; 𝜉, 𝜂 𝑑𝜉𝑑𝜂
−∞ −∞
Shift Invariance System
A system is shift invariant if an arbitrary translation of
the input results in an identical translation in the output.
Then with linearity we can know the system response
to any input.
Let the input
𝑓𝑥0 𝑦0 𝑥, 𝑦 = 𝑓 𝑥 − 𝑥0 , 𝑦 − 𝑦0
then the output g(𝑥 − 𝑥0 , 𝑦 − 𝑦0 )= Ϩ [𝑓𝑥0 𝑦0 𝑥, 𝑦 ]
System response to a shifted impulse
Ϩ [𝛿𝜉𝜂 𝑥, 𝑦 ]= h(𝑥 − 𝜉, 𝑦 − 𝜂)
Linear Shift-Invariance (LSI) System
Linear shift-invariant (LSI) System Response
∞
∞
𝑔 𝑥, 𝑦 =
𝑓 𝜉, 𝜂 ℎ 𝑥 − 𝜉, 𝑦 − 𝜂 𝑑𝜉𝑑𝜂
−∞ −∞
Convolution Integral representation of system response
𝑔 𝑥, 𝑦 = ℎ 𝑥, 𝑦 ∗ 𝑓(𝑥, 𝑦)
Example: Consider a continuous system with inputoutput equation g(x,y) = xyf(x,y).
Is the system linear and shift-invariant?
Connection of LSI Systems
Cascade
Parallel
𝑔 𝑥, 𝑦 = ℎ1 𝑥, 𝑦 ∗ ℎ2 𝑥, 𝑦 ∗ 𝑓(𝑥, 𝑦)
𝑔 𝑥, 𝑦 = [ℎ1 𝑥, 𝑦 + ℎ2 𝑥, 𝑦 ] ∗ 𝑓(𝑥, 𝑦)
Connection of LSI Systems
Example: Consider two LSI systems connected in
cascade, with Gaussian PSFs of the form:
1
−
ℎ1 𝑥, 𝑦 =
𝑒
2𝜋𝜎12
𝑥 2 +𝑦 2 /2𝜎12
1
−
ℎ2 𝑥, 𝑦 =
𝑒
2𝜋𝜎22
𝑥 2 +𝑦 2 /2𝜎22
where σ1 and σ2 are two positive constants.
What is the PSF of the system?
Separable Systems
A 2-D LSI system with PSF h(x, y) is a separable
system if there are two 1-D systems with PSFs h1(x)
and h2(x), such that h(x,y) = h1(x)h2(x)
1
−
ℎ 𝑥, 𝑦 =
𝑒
2𝜋𝜎 2
𝑥 2 +𝑦 2 /2𝜎 2
This PSF is separable
1
2 /2𝜎 2
−𝑥
ℎ1 𝑥 =
𝑒
2𝜋𝜎
1
2 /2𝜎 2
−𝑦
ℎ2 𝑥 =
𝑒
2𝜋𝜎
Separable Systems
In practice it is easier and faster to execute two
consecutive 1-D operations than a single 2-D operation.
∞
𝑤 𝑥, 𝑦 =
𝑓 𝜉, 𝑦 ℎ1 𝑥 − 𝜉 𝑑𝜉
−∞
g 𝑥, 𝑦 =
For every y
∞
𝑤 𝑥, 𝜂 ℎ2 𝑦 − 𝜂 𝑑𝜂
−∞
For every x
Stable Systems
A system is a bounded-input bounded-output (BIBO)
stable system if
For bounded input
|𝑓 𝑥, 𝑦 | ≤ 𝐵 < ∞ for every (x, y)
The output is bounded
𝑔(𝑥, 𝑦) = ℎ 𝑥, 𝑦 ∗ 𝑓(𝑥, 𝑦) ≤ 𝐵, < ∞
and
∞
∞
|ℎ 𝑥, 𝑦 |𝑑𝑥𝑑𝑦 < ∞
−∞ −∞
1-D Fourier Transform (time)
Continuous 1-D Fourier Transform

 j 2ft
x
(
t
)
e
dt

X (2f ) 

Discrete 1-D Fourier Transform
N -1
X ( k )   x ( n )e
j
2k
n
N
n 0
x(n)
X(k)
= [125
= [668
|X(k)| = [668
Phase = [0
145
-29.2 - j38
47.9
-127.5
148
7.7 - j12.96
15.1
-59.3
140
7.7 - j12.96
15.1
59.3
110]
-29.2 - j38]
47.9]
127.5]
1-D Fourier Transform
1-D Fourier transform
∞
𝐹 𝑢 = ℱ1𝐷 𝑓 𝑢 =
𝑓 𝑥 𝑒 −𝑗2𝜋𝑢𝑥 𝑑𝑥
−∞
u is the spatial frequency
1-D inverse Fourier transform
−1
𝑓 𝑥 = ℱ1𝐷
𝐹 𝑥 =
Example:
∞
𝐹 𝑢 𝑒 𝑗2𝜋𝑢𝑥 𝑑𝑢
−∞
What is the Fourier
1,
transform of the 𝑟𝑒𝑐𝑡(𝑥) =
0,
1
for |𝑥| <
2
1
for |𝑥| >
2
Fourier Transform
The 2-D Fourier transform of f(x, y)
∞
𝐹 𝑢, 𝑣 = ℱ2𝐷 𝑓 𝑢, 𝑣 =
∞
𝑓 𝑥, 𝑦 𝑒 −𝑗2𝜋(𝑢𝑥+𝑣𝑦) 𝑑𝑥𝑑𝑦
−∞ −∞
u and v are the spatial frequencies
The 2-D inverse Fourier transform of F(u, v)
∞
𝑓(𝑥, 𝑦) =
∞
𝐹 𝑢, 𝑣 𝑒 𝑗2𝜋(𝑢𝑥+𝑣𝑦) 𝑑𝑢𝑑𝑣
−∞ −∞
Fourier Transform
Magnitude (magnitude spectrum) of FT
𝐹(𝑢, 𝑣) =
𝐹𝑅2 𝑢, 𝑣 + 𝐹𝐼2 𝑢, 𝑣
Angle (phase spectrum) of the FT
∠𝐹 𝑢, 𝑣 = 𝑡𝑎𝑛
−1
𝐹𝐼 (𝑢, 𝑣)
𝐹𝑅 (𝑢, 𝑣)
𝐹(𝑢, 𝑣) = 𝐹(𝑢, 𝑣) 𝑒 𝑗∠𝐹
𝑢,𝑣
Example: What is the Fourier transform of the point
impulse 𝛿(𝑥, 𝑦)?
Fourier Transform Pairs
Examples of Fourier Transform
Example: What is the Fourier transform of
𝑓 𝑥, 𝑦 = 𝑒 𝑗2𝜋(𝑢0 𝑥+𝑣0𝑦)
Answer:
ℱ2𝐷 𝑓 𝑢, 𝑣 = 𝛿 𝑢 − 𝑢0 , 𝑣 − 𝑣0
If the spatial frequency u0 and v0 are zero then f(x,y) =1
and the spectrum F(u,v) will be 𝛿 𝑢, 𝑣 .
Slow signal variation in space produces a spectral
content that is primarily concentrated at low
frequencies.
Examples of Fourier Transform
Three images of decreasing spatial variation (from left to right) and the
associated magnitude spectra [depicted as log(1 + |F(u, υ)|)].
Examples of Fourier Transform
>> img1 = imread('\\PHYSICSSERVER\MBingabr\BiomedicalImaging\mri.tif');
>> imshow(img1)
>> size(img1)
ans = 256 256
>> FFT_img1 = fftshift(fft2(img1));
>> Abs_FFT_img1 = abs(FFT_img1)
>> surf(Abs_FFT_img1(110:140,110:140))
>> Log_Abs_FFT_img1=log10(1+Abs_FFT_img1);
>> surf(Log_Abs_FFT_img1(110:140,110:140))
Properties of the Fourier Transform
Properties are used in theory and application to
simplify calculation.
Linearity
ℱ2𝐷 𝑎1 𝑓 + 𝑎2 𝑔 𝑢, 𝑣 = 𝑎1 𝐹 𝑢, 𝑣 + 𝑎2 𝐺(𝑢, 𝑣)
Translation
If F(u,v) is the FT of a signal f(x, y) then the FT
of a translated signal
𝑓𝑥0 𝑦0 𝑥, 𝑦 = 𝑓 𝑥 − 𝑥0 , 𝑦 − 𝑦0
is
ℱ2𝐷 𝑓𝑥0 𝑦0 𝑢, 𝑣 = 𝐹 𝑢, 𝑣 𝑒 −𝑗2𝜋(𝑢𝑥0 +𝑣𝑦0)
Properties of the Fourier Transform
Conjugation and Conjugate Symmetry
If F(u,v) is the FT of a signal f(x, y) then
𝐹(𝑢, 𝑣) = 𝐹 ∗ (−𝑢, −𝑣)
𝐹𝑅 (𝑢, 𝑣) = 𝐹𝑅 (−𝑢, −𝑣)
|𝐹 𝑢, 𝑣 | = |𝐹 −𝑢, −𝑣 |
∠𝐹 𝑢, 𝑣 = −∠𝐹 −𝑢, −𝑣
𝐹𝐼 (𝑢, 𝑣) = −𝐹𝐼 (−𝑢, −𝑣)
Properties of the Fourier Transform
Scaling
If F(u,v) is the FT of a signal f(x, y) and if
𝑓𝑎𝑏 𝑥, 𝑦 = 𝑓 𝑎𝑥, 𝑏𝑦
1
𝑢 𝑣
ℱ2𝐷 𝑓𝑎𝑏 𝑢, 𝑣 =
𝐹 ,
|𝑎𝑏| 𝑎 𝑏
Example
Detectors of many medical imaging systems can be
modeled as rect functions of different sizes and
locations. Compute the FT of the following
𝑥 − 𝑥0 𝑦 − 𝑦0
𝑓 𝑥, 𝑦 = 𝑟𝑒𝑐𝑡
,
∆𝑥
∆𝑦
Properties of the Fourier Transform
Rotation
If F(u,v) is the FT of a signal f(x, y) and if
𝑓𝜃 𝑥, 𝑦 = 𝑓 𝑥𝑐𝑜𝑠𝜃 − 𝑦𝑠𝑖𝑛𝜃, 𝑥𝑠𝑖𝑛𝜃 + 𝑦𝑐𝑜𝑠𝜃
ℱ2𝐷 𝑓𝜃 𝑢, 𝑣 = 𝐹 𝑢 𝑐𝑜𝑠𝜃 − 𝑣 𝑠𝑖𝑛𝜃, 𝑢 𝑠𝑖𝑛𝜃 + 𝑣 𝑐𝑜𝑠𝜃
If f(x, y) is rotated by an angle 𝜃, then its FT is rotated by
the same angle.
Properties of the Fourier Transform
Convolution
The Fourier transform of the convolution
f(x, y) * g(x, y) is
ℱ2𝐷 𝑓 ∗ 𝑔 𝑢, 𝑣 = 𝐹 𝑢 , 𝑣 𝐺 𝑢 , 𝑣
Convolution property simplify the difficult task of calculating
the convolution in the spatial domain to multiplication in the
frequency domain.
Example:
Find Fourier transform of the convolution f(x, y) * g(x, y)
𝑔 𝑥, 𝑦 = 𝑠𝑖𝑛𝑐 𝑉𝑥, 𝑈𝑦
𝑓 𝑥, 𝑦 = 𝑠𝑖𝑛𝑐 𝑈𝑥, 𝑉𝑦
0<VU
Properties of the Fourier Transform
Product
The Fourier transform of the product f(x, y) g(x, y) is the
convolution of their Fourier transforms.
ℱ2𝐷 𝑓𝑔 𝑢, 𝑣 = 𝐹 𝑢 , 𝑣 ∗ 𝐺 𝑢 , 𝑣
∞
=
∞
𝐺 𝜉, 𝜂 𝐹 𝑢 − 𝜉, 𝑣 − 𝜂 𝑑𝜉𝑑𝜂
−∞ −∞
Separable Product
If f(x, y)=f1(x)f2(y) then ℱ2𝐷 𝑓 𝑢, 𝑣 = 𝐹1 (𝑢)𝐹2 (𝑣)
where
𝐹1 𝑢 = ℱ1𝐷 𝑓1 𝑢
Separability of the Fourier Transform
The Fourier transform F(u,v) of a 2-D signal f(x, y) can be
calculated using two simpler 1-D Fourier transforms, as
follows:
∞
1) 𝑟 𝑢, 𝑦 =
𝑓 𝑥, 𝑦 𝑒 −𝑗2𝜋𝑢𝑥 𝑑𝑥
−∞
∞
2) 𝐹 𝑢, 𝑦 =
−∞
𝑟 𝑢, 𝑦 𝑒 −𝑗2𝜋𝑢𝑦 𝑑𝑦
For every y.
For every x.
Transfer Function
The system’s transfer function (frequency response) H(u, v)
is the Fourier transform of the system’s PSF h(x,y).
∞
∞
𝐻 𝑢, 𝑣 =
ℎ 𝜉, 𝜂 𝑒 −𝑗2𝜋(𝑢𝜉+𝑣𝜂) 𝑑𝜉𝑑𝜂
−∞ −∞
The inverse Fourier transform of the transfer function H(u, v)
is the point spread function h(x,y).
∞
∞
ℎ 𝑥, 𝑦 =
𝐻 𝑢, 𝑣 𝑒 −𝑗2𝜋(𝑢𝑥+𝑣𝑦) 𝑑𝑢𝑑𝑣
−∞ −∞
The output G(u, v) of a system in response to input F(u, v) is
the product of the input with the transfer function H(u, v) .
𝐺 𝑢, 𝑣 = 𝐻 𝑢 , 𝑣 𝐹 𝑢 , 𝑣
Transfer Function
Example: Consider an idealized system whose PSF is
h(x,y) = (x-x0, y-y0). What is the transfer function H(u, v) of
the system, and what is the system output g(x,y) to an input
signal f(x,y).
Low Pass Filter
𝐻(𝑢, 𝑣) =
𝐺(𝑢, 𝑣) =
1,
for 𝑢2 + 𝑣 2 ≤ 𝑐
0,
for 𝑢2 + 𝑣 2 > 𝑐
𝐹(𝑢, 𝑣) for 𝑢2 + 𝑣 2 ≤ 𝑐
0,
for 𝑢2 + 𝑣 2 > 𝑐
c1 > c2
Circular Symmetry
Often, the performance of a medical imaging system does
not depend on the orientation of the patient with respect to
the system. The independence arises from the circular
symmetry of the PSF.
A 2-D signal f(x, y) is circularly symmetric if
fθ(x, y) = f(x, y) for every θ.
Property of Circular Symmetry
• f(x, y) is even in both x and y
• F(u, v) is even in both u and v
• | F(u, v) | = F(u, v)
• ∠F(u, v) = 0
• f(x, y) = f(r) where 𝑟 =
𝑥2 + 𝑦2
• F(u, v) = F(q) where 𝑞 =
𝑢2 + 𝑣 2
f(r) and F(q) are one dimensional signals representing two
dimensional signals
Hankel Transform
The relationship between f(r) and F(q) is determined by
Hankel Transform.
∞
𝐹 𝑞 = 2𝜋
0
𝑓 𝑟 𝐽𝑜 2𝜋𝑞𝑟 𝑟𝑑𝑟
𝐹 𝑞 = ℋ 𝑓(𝑟)
where J0(r) is the zero-order Bessel function of the first kind.
1 𝜋
𝐽0 𝑟 =
cos 𝑟𝑠𝑖𝑛𝜙 𝑑𝜙
𝜋 0
The nth-order Bessel function
1 𝜋
𝐽𝑛 𝑟 =
cos 𝑛𝜙 − 𝑟𝑠𝑖𝑛𝜙 𝑑𝜙
𝜋 0
for n = 0, 1, 2, …
Hankel Transform
The inverse Hankel transform.
∞
𝑓 𝑟 = 2𝜋
0
unit disk
𝐹 𝑞 𝐽𝑜 2𝜋𝑞𝑟 𝑞𝑑𝑞
jink function
sinc and jink functions
Example
In some medical imaging systems, only spatial frequencies
smaller than q0 can be imaged. What is the function having
uniform spatial frequencies within the desk of radius q0 and
what is its inverse Fourier transform.